sleeping-ai/suno-dataset

Excited to share SUNO Collection, a corpus of 100,753 songs and its respective videos. We are the largest known valid collection of the entire SUNO database till 3rd of July, 2025. Previous, nyuuzyou published a suno dataset consisting of 690K songs, user profiles and data points.

That dataset while a good contribution, it was riddled with duplicated user handles and same songs over and over. It was a naïve attempt to provide a database. We have went through the entire contribution of nyuuzyou, and found following:

1. After dedup of user_handles there were approx. 120K valid user handles
2. We ran a second round of scraping to find new users and found 13K fresh users that nyuuzyou had missed.
3. Finally, we run a program to separate valid userhandles Vs dummy accounts and found 8,558 user profiles that were either valid or had a few songs and followers.

We did the entire data collection process basing foundation on nyuuzyou's HF contribution and found 100, 753 unique songs on SUNO platform till this date. We also estimate that our entire data-collection had no more than 3% loss.

Why this dataset?

SUNO is the most admired artificial intelligence song creator platform and has a strong community behind it. Provided that artificially created songs are gaining popularity both commercially and among researchers, we believe this dataset will help the open-source community and researchers to study them and learn about these models that are still in early phase but the community is liking its quality.

Besides, there had been papers at ICLR 2025 and SCIENCE journal that were based on SUNO.

1. SONICS
2. SunoCaps

We simply wanted to provide a better dataset and a complete contribution that wasn't shrouded under Owh! Yah, they provide a synthetic dataset based on scraping, Fine, even if the quality is poor.

Our entire motive was to be the most perfect contribution.

We have used responsible scraping techniques and made sure, we didn't overload their servers.

What data fields we provide?

user_id: tracing the actual user on SUNO platform through this ID. audio_url: Link to download audio video_url: Link to download video model_name: indicating which SUNO model had been used. major_model_version: indicating the SUNO model version number

Ethics statement

Sleeping AI didn't do anything different than what already exists as open-source contributions on Huggingface and Kaggle. We do value individual privacy and limit on who can receive the entire SUNO dump that includes lyrics, prompts and much more than what we are posting here.

Community Call

We invite individuals interested in helping us create the largest known artificially generated song corpus for training and studying their impact on the Generative Music Modelling community and we plan to write a technical report/an interesting full-paper on arXiv on that.

LICENCE

We use CC-by-nc-nd 4.0, which restricts the redistribution and making derivatives of this dataset to be shared on the internet without the permission of Sleeping AI.

Feel free to reach out on Twitter(now X): https://x.com/sleeping4cat